xref: /openbmc/linux/fs/inode.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /*
2  * linux/fs/inode.c
3  *
4  * (C) 1997 Linus Torvalds
5  */
6 
7 #include <linux/config.h>
8 #include <linux/fs.h>
9 #include <linux/mm.h>
10 #include <linux/dcache.h>
11 #include <linux/init.h>
12 #include <linux/quotaops.h>
13 #include <linux/slab.h>
14 #include <linux/writeback.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/hash.h>
19 #include <linux/swap.h>
20 #include <linux/security.h>
21 #include <linux/pagemap.h>
22 #include <linux/cdev.h>
23 #include <linux/bootmem.h>
24 #include <linux/inotify.h>
25 
26 /*
27  * This is needed for the following functions:
28  *  - inode_has_buffers
29  *  - invalidate_inode_buffers
30  *  - invalidate_bdev
31  *
32  * FIXME: remove all knowledge of the buffer layer from this file
33  */
34 #include <linux/buffer_head.h>
35 
36 /*
37  * New inode.c implementation.
38  *
39  * This implementation has the basic premise of trying
40  * to be extremely low-overhead and SMP-safe, yet be
41  * simple enough to be "obviously correct".
42  *
43  * Famous last words.
44  */
45 
46 /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */
47 
48 /* #define INODE_PARANOIA 1 */
49 /* #define INODE_DEBUG 1 */
50 
51 /*
52  * Inode lookup is no longer as critical as it used to be:
53  * most of the lookups are going to be through the dcache.
54  */
55 #define I_HASHBITS	i_hash_shift
56 #define I_HASHMASK	i_hash_mask
57 
58 static unsigned int i_hash_mask;
59 static unsigned int i_hash_shift;
60 
61 /*
62  * Each inode can be on two separate lists. One is
63  * the hash list of the inode, used for lookups. The
64  * other linked list is the "type" list:
65  *  "in_use" - valid inode, i_count > 0, i_nlink > 0
66  *  "dirty"  - as "in_use" but also dirty
67  *  "unused" - valid inode, i_count = 0
68  *
69  * A "dirty" list is maintained for each super block,
70  * allowing for low-overhead inode sync() operations.
71  */
72 
73 LIST_HEAD(inode_in_use);
74 LIST_HEAD(inode_unused);
75 static struct hlist_head *inode_hashtable;
76 
77 /*
78  * A simple spinlock to protect the list manipulations.
79  *
80  * NOTE! You also have to own the lock if you change
81  * the i_state of an inode while it is in use..
82  */
83 DEFINE_SPINLOCK(inode_lock);
84 
85 /*
86  * iprune_sem provides exclusion between the kswapd or try_to_free_pages
87  * icache shrinking path, and the umount path.  Without this exclusion,
88  * by the time prune_icache calls iput for the inode whose pages it has
89  * been invalidating, or by the time it calls clear_inode & destroy_inode
90  * from its final dispose_list, the struct super_block they refer to
91  * (for inode->i_sb->s_op) may already have been freed and reused.
92  */
93 DECLARE_MUTEX(iprune_sem);
94 
95 /*
96  * Statistics gathering..
97  */
98 struct inodes_stat_t inodes_stat;
99 
100 static kmem_cache_t * inode_cachep;
101 
102 static struct inode *alloc_inode(struct super_block *sb)
103 {
104 	static struct address_space_operations empty_aops;
105 	static struct inode_operations empty_iops;
106 	static struct file_operations empty_fops;
107 	struct inode *inode;
108 
109 	if (sb->s_op->alloc_inode)
110 		inode = sb->s_op->alloc_inode(sb);
111 	else
112 		inode = (struct inode *) kmem_cache_alloc(inode_cachep, SLAB_KERNEL);
113 
114 	if (inode) {
115 		struct address_space * const mapping = &inode->i_data;
116 
117 		inode->i_sb = sb;
118 		inode->i_blkbits = sb->s_blocksize_bits;
119 		inode->i_flags = 0;
120 		atomic_set(&inode->i_count, 1);
121 		inode->i_op = &empty_iops;
122 		inode->i_fop = &empty_fops;
123 		inode->i_nlink = 1;
124 		atomic_set(&inode->i_writecount, 0);
125 		inode->i_size = 0;
126 		inode->i_blocks = 0;
127 		inode->i_bytes = 0;
128 		inode->i_generation = 0;
129 #ifdef CONFIG_QUOTA
130 		memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
131 #endif
132 		inode->i_pipe = NULL;
133 		inode->i_bdev = NULL;
134 		inode->i_cdev = NULL;
135 		inode->i_rdev = 0;
136 		inode->i_security = NULL;
137 		inode->dirtied_when = 0;
138 		if (security_inode_alloc(inode)) {
139 			if (inode->i_sb->s_op->destroy_inode)
140 				inode->i_sb->s_op->destroy_inode(inode);
141 			else
142 				kmem_cache_free(inode_cachep, (inode));
143 			return NULL;
144 		}
145 
146 		mapping->a_ops = &empty_aops;
147  		mapping->host = inode;
148 		mapping->flags = 0;
149 		mapping_set_gfp_mask(mapping, GFP_HIGHUSER);
150 		mapping->assoc_mapping = NULL;
151 		mapping->backing_dev_info = &default_backing_dev_info;
152 
153 		/*
154 		 * If the block_device provides a backing_dev_info for client
155 		 * inodes then use that.  Otherwise the inode share the bdev's
156 		 * backing_dev_info.
157 		 */
158 		if (sb->s_bdev) {
159 			struct backing_dev_info *bdi;
160 
161 			bdi = sb->s_bdev->bd_inode_backing_dev_info;
162 			if (!bdi)
163 				bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
164 			mapping->backing_dev_info = bdi;
165 		}
166 		memset(&inode->u, 0, sizeof(inode->u));
167 		inode->i_mapping = mapping;
168 	}
169 	return inode;
170 }
171 
172 void destroy_inode(struct inode *inode)
173 {
174 	if (inode_has_buffers(inode))
175 		BUG();
176 	security_inode_free(inode);
177 	if (inode->i_sb->s_op->destroy_inode)
178 		inode->i_sb->s_op->destroy_inode(inode);
179 	else
180 		kmem_cache_free(inode_cachep, (inode));
181 }
182 
183 
184 /*
185  * These are initializations that only need to be done
186  * once, because the fields are idempotent across use
187  * of the inode, so let the slab aware of that.
188  */
189 void inode_init_once(struct inode *inode)
190 {
191 	memset(inode, 0, sizeof(*inode));
192 	INIT_HLIST_NODE(&inode->i_hash);
193 	INIT_LIST_HEAD(&inode->i_dentry);
194 	INIT_LIST_HEAD(&inode->i_devices);
195 	sema_init(&inode->i_sem, 1);
196 	init_rwsem(&inode->i_alloc_sem);
197 	INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC);
198 	rwlock_init(&inode->i_data.tree_lock);
199 	spin_lock_init(&inode->i_data.i_mmap_lock);
200 	INIT_LIST_HEAD(&inode->i_data.private_list);
201 	spin_lock_init(&inode->i_data.private_lock);
202 	INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
203 	INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
204 	spin_lock_init(&inode->i_lock);
205 	i_size_ordered_init(inode);
206 #ifdef CONFIG_INOTIFY
207 	INIT_LIST_HEAD(&inode->inotify_watches);
208 	sema_init(&inode->inotify_sem, 1);
209 #endif
210 }
211 
212 EXPORT_SYMBOL(inode_init_once);
213 
214 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
215 {
216 	struct inode * inode = (struct inode *) foo;
217 
218 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
219 	    SLAB_CTOR_CONSTRUCTOR)
220 		inode_init_once(inode);
221 }
222 
223 /*
224  * inode_lock must be held
225  */
226 void __iget(struct inode * inode)
227 {
228 	if (atomic_read(&inode->i_count)) {
229 		atomic_inc(&inode->i_count);
230 		return;
231 	}
232 	atomic_inc(&inode->i_count);
233 	if (!(inode->i_state & (I_DIRTY|I_LOCK)))
234 		list_move(&inode->i_list, &inode_in_use);
235 	inodes_stat.nr_unused--;
236 }
237 
238 /**
239  * clear_inode - clear an inode
240  * @inode: inode to clear
241  *
242  * This is called by the filesystem to tell us
243  * that the inode is no longer useful. We just
244  * terminate it with extreme prejudice.
245  */
246 void clear_inode(struct inode *inode)
247 {
248 	might_sleep();
249 	invalidate_inode_buffers(inode);
250 
251 	if (inode->i_data.nrpages)
252 		BUG();
253 	if (!(inode->i_state & I_FREEING))
254 		BUG();
255 	if (inode->i_state & I_CLEAR)
256 		BUG();
257 	wait_on_inode(inode);
258 	DQUOT_DROP(inode);
259 	if (inode->i_sb && inode->i_sb->s_op->clear_inode)
260 		inode->i_sb->s_op->clear_inode(inode);
261 	if (inode->i_bdev)
262 		bd_forget(inode);
263 	if (inode->i_cdev)
264 		cd_forget(inode);
265 	inode->i_state = I_CLEAR;
266 }
267 
268 EXPORT_SYMBOL(clear_inode);
269 
270 /*
271  * dispose_list - dispose of the contents of a local list
272  * @head: the head of the list to free
273  *
274  * Dispose-list gets a local list with local inodes in it, so it doesn't
275  * need to worry about list corruption and SMP locks.
276  */
277 static void dispose_list(struct list_head *head)
278 {
279 	int nr_disposed = 0;
280 
281 	while (!list_empty(head)) {
282 		struct inode *inode;
283 
284 		inode = list_entry(head->next, struct inode, i_list);
285 		list_del(&inode->i_list);
286 
287 		if (inode->i_data.nrpages)
288 			truncate_inode_pages(&inode->i_data, 0);
289 		clear_inode(inode);
290 
291 		spin_lock(&inode_lock);
292 		hlist_del_init(&inode->i_hash);
293 		list_del_init(&inode->i_sb_list);
294 		spin_unlock(&inode_lock);
295 
296 		wake_up_inode(inode);
297 		destroy_inode(inode);
298 		nr_disposed++;
299 	}
300 	spin_lock(&inode_lock);
301 	inodes_stat.nr_inodes -= nr_disposed;
302 	spin_unlock(&inode_lock);
303 }
304 
305 /*
306  * Invalidate all inodes for a device.
307  */
308 static int invalidate_list(struct list_head *head, struct list_head *dispose)
309 {
310 	struct list_head *next;
311 	int busy = 0, count = 0;
312 
313 	next = head->next;
314 	for (;;) {
315 		struct list_head * tmp = next;
316 		struct inode * inode;
317 
318 		/*
319 		 * We can reschedule here without worrying about the list's
320 		 * consistency because the per-sb list of inodes must not
321 		 * change during umount anymore, and because iprune_sem keeps
322 		 * shrink_icache_memory() away.
323 		 */
324 		cond_resched_lock(&inode_lock);
325 
326 		next = next->next;
327 		if (tmp == head)
328 			break;
329 		inode = list_entry(tmp, struct inode, i_sb_list);
330 		invalidate_inode_buffers(inode);
331 		if (!atomic_read(&inode->i_count)) {
332 			list_move(&inode->i_list, dispose);
333 			inode->i_state |= I_FREEING;
334 			count++;
335 			continue;
336 		}
337 		busy = 1;
338 	}
339 	/* only unused inodes may be cached with i_count zero */
340 	inodes_stat.nr_unused -= count;
341 	return busy;
342 }
343 
344 /**
345  *	invalidate_inodes	- discard the inodes on a device
346  *	@sb: superblock
347  *
348  *	Discard all of the inodes for a given superblock. If the discard
349  *	fails because there are busy inodes then a non zero value is returned.
350  *	If the discard is successful all the inodes have been discarded.
351  */
352 int invalidate_inodes(struct super_block * sb)
353 {
354 	int busy;
355 	LIST_HEAD(throw_away);
356 
357 	down(&iprune_sem);
358 	spin_lock(&inode_lock);
359 	inotify_unmount_inodes(&sb->s_inodes);
360 	busy = invalidate_list(&sb->s_inodes, &throw_away);
361 	spin_unlock(&inode_lock);
362 
363 	dispose_list(&throw_away);
364 	up(&iprune_sem);
365 
366 	return busy;
367 }
368 
369 EXPORT_SYMBOL(invalidate_inodes);
370 
371 int __invalidate_device(struct block_device *bdev)
372 {
373 	struct super_block *sb = get_super(bdev);
374 	int res = 0;
375 
376 	if (sb) {
377 		/*
378 		 * no need to lock the super, get_super holds the
379 		 * read semaphore so the filesystem cannot go away
380 		 * under us (->put_super runs with the write lock
381 		 * hold).
382 		 */
383 		shrink_dcache_sb(sb);
384 		res = invalidate_inodes(sb);
385 		drop_super(sb);
386 	}
387 	invalidate_bdev(bdev, 0);
388 	return res;
389 }
390 EXPORT_SYMBOL(__invalidate_device);
391 
392 static int can_unuse(struct inode *inode)
393 {
394 	if (inode->i_state)
395 		return 0;
396 	if (inode_has_buffers(inode))
397 		return 0;
398 	if (atomic_read(&inode->i_count))
399 		return 0;
400 	if (inode->i_data.nrpages)
401 		return 0;
402 	return 1;
403 }
404 
405 /*
406  * Scan `goal' inodes on the unused list for freeable ones. They are moved to
407  * a temporary list and then are freed outside inode_lock by dispose_list().
408  *
409  * Any inodes which are pinned purely because of attached pagecache have their
410  * pagecache removed.  We expect the final iput() on that inode to add it to
411  * the front of the inode_unused list.  So look for it there and if the
412  * inode is still freeable, proceed.  The right inode is found 99.9% of the
413  * time in testing on a 4-way.
414  *
415  * If the inode has metadata buffers attached to mapping->private_list then
416  * try to remove them.
417  */
418 static void prune_icache(int nr_to_scan)
419 {
420 	LIST_HEAD(freeable);
421 	int nr_pruned = 0;
422 	int nr_scanned;
423 	unsigned long reap = 0;
424 
425 	down(&iprune_sem);
426 	spin_lock(&inode_lock);
427 	for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) {
428 		struct inode *inode;
429 
430 		if (list_empty(&inode_unused))
431 			break;
432 
433 		inode = list_entry(inode_unused.prev, struct inode, i_list);
434 
435 		if (inode->i_state || atomic_read(&inode->i_count)) {
436 			list_move(&inode->i_list, &inode_unused);
437 			continue;
438 		}
439 		if (inode_has_buffers(inode) || inode->i_data.nrpages) {
440 			__iget(inode);
441 			spin_unlock(&inode_lock);
442 			if (remove_inode_buffers(inode))
443 				reap += invalidate_inode_pages(&inode->i_data);
444 			iput(inode);
445 			spin_lock(&inode_lock);
446 
447 			if (inode != list_entry(inode_unused.next,
448 						struct inode, i_list))
449 				continue;	/* wrong inode or list_empty */
450 			if (!can_unuse(inode))
451 				continue;
452 		}
453 		list_move(&inode->i_list, &freeable);
454 		inode->i_state |= I_FREEING;
455 		nr_pruned++;
456 	}
457 	inodes_stat.nr_unused -= nr_pruned;
458 	spin_unlock(&inode_lock);
459 
460 	dispose_list(&freeable);
461 	up(&iprune_sem);
462 
463 	if (current_is_kswapd())
464 		mod_page_state(kswapd_inodesteal, reap);
465 	else
466 		mod_page_state(pginodesteal, reap);
467 }
468 
469 /*
470  * shrink_icache_memory() will attempt to reclaim some unused inodes.  Here,
471  * "unused" means that no dentries are referring to the inodes: the files are
472  * not open and the dcache references to those inodes have already been
473  * reclaimed.
474  *
475  * This function is passed the number of inodes to scan, and it returns the
476  * total number of remaining possibly-reclaimable inodes.
477  */
478 static int shrink_icache_memory(int nr, gfp_t gfp_mask)
479 {
480 	if (nr) {
481 		/*
482 		 * Nasty deadlock avoidance.  We may hold various FS locks,
483 		 * and we don't want to recurse into the FS that called us
484 		 * in clear_inode() and friends..
485 	 	 */
486 		if (!(gfp_mask & __GFP_FS))
487 			return -1;
488 		prune_icache(nr);
489 	}
490 	return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
491 }
492 
493 static void __wait_on_freeing_inode(struct inode *inode);
494 /*
495  * Called with the inode lock held.
496  * NOTE: we are not increasing the inode-refcount, you must call __iget()
497  * by hand after calling find_inode now! This simplifies iunique and won't
498  * add any additional branch in the common code.
499  */
500 static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data)
501 {
502 	struct hlist_node *node;
503 	struct inode * inode = NULL;
504 
505 repeat:
506 	hlist_for_each (node, head) {
507 		inode = hlist_entry(node, struct inode, i_hash);
508 		if (inode->i_sb != sb)
509 			continue;
510 		if (!test(inode, data))
511 			continue;
512 		if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
513 			__wait_on_freeing_inode(inode);
514 			goto repeat;
515 		}
516 		break;
517 	}
518 	return node ? inode : NULL;
519 }
520 
521 /*
522  * find_inode_fast is the fast path version of find_inode, see the comment at
523  * iget_locked for details.
524  */
525 static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino)
526 {
527 	struct hlist_node *node;
528 	struct inode * inode = NULL;
529 
530 repeat:
531 	hlist_for_each (node, head) {
532 		inode = hlist_entry(node, struct inode, i_hash);
533 		if (inode->i_ino != ino)
534 			continue;
535 		if (inode->i_sb != sb)
536 			continue;
537 		if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) {
538 			__wait_on_freeing_inode(inode);
539 			goto repeat;
540 		}
541 		break;
542 	}
543 	return node ? inode : NULL;
544 }
545 
546 /**
547  *	new_inode 	- obtain an inode
548  *	@sb: superblock
549  *
550  *	Allocates a new inode for given superblock.
551  */
552 struct inode *new_inode(struct super_block *sb)
553 {
554 	static unsigned long last_ino;
555 	struct inode * inode;
556 
557 	spin_lock_prefetch(&inode_lock);
558 
559 	inode = alloc_inode(sb);
560 	if (inode) {
561 		spin_lock(&inode_lock);
562 		inodes_stat.nr_inodes++;
563 		list_add(&inode->i_list, &inode_in_use);
564 		list_add(&inode->i_sb_list, &sb->s_inodes);
565 		inode->i_ino = ++last_ino;
566 		inode->i_state = 0;
567 		spin_unlock(&inode_lock);
568 	}
569 	return inode;
570 }
571 
572 EXPORT_SYMBOL(new_inode);
573 
574 void unlock_new_inode(struct inode *inode)
575 {
576 	/*
577 	 * This is special!  We do not need the spinlock
578 	 * when clearing I_LOCK, because we're guaranteed
579 	 * that nobody else tries to do anything about the
580 	 * state of the inode when it is locked, as we
581 	 * just created it (so there can be no old holders
582 	 * that haven't tested I_LOCK).
583 	 */
584 	inode->i_state &= ~(I_LOCK|I_NEW);
585 	wake_up_inode(inode);
586 }
587 
588 EXPORT_SYMBOL(unlock_new_inode);
589 
590 /*
591  * This is called without the inode lock held.. Be careful.
592  *
593  * We no longer cache the sb_flags in i_flags - see fs.h
594  *	-- rmk@arm.uk.linux.org
595  */
596 static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data)
597 {
598 	struct inode * inode;
599 
600 	inode = alloc_inode(sb);
601 	if (inode) {
602 		struct inode * old;
603 
604 		spin_lock(&inode_lock);
605 		/* We released the lock, so.. */
606 		old = find_inode(sb, head, test, data);
607 		if (!old) {
608 			if (set(inode, data))
609 				goto set_failed;
610 
611 			inodes_stat.nr_inodes++;
612 			list_add(&inode->i_list, &inode_in_use);
613 			list_add(&inode->i_sb_list, &sb->s_inodes);
614 			hlist_add_head(&inode->i_hash, head);
615 			inode->i_state = I_LOCK|I_NEW;
616 			spin_unlock(&inode_lock);
617 
618 			/* Return the locked inode with I_NEW set, the
619 			 * caller is responsible for filling in the contents
620 			 */
621 			return inode;
622 		}
623 
624 		/*
625 		 * Uhhuh, somebody else created the same inode under
626 		 * us. Use the old inode instead of the one we just
627 		 * allocated.
628 		 */
629 		__iget(old);
630 		spin_unlock(&inode_lock);
631 		destroy_inode(inode);
632 		inode = old;
633 		wait_on_inode(inode);
634 	}
635 	return inode;
636 
637 set_failed:
638 	spin_unlock(&inode_lock);
639 	destroy_inode(inode);
640 	return NULL;
641 }
642 
643 /*
644  * get_new_inode_fast is the fast path version of get_new_inode, see the
645  * comment at iget_locked for details.
646  */
647 static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino)
648 {
649 	struct inode * inode;
650 
651 	inode = alloc_inode(sb);
652 	if (inode) {
653 		struct inode * old;
654 
655 		spin_lock(&inode_lock);
656 		/* We released the lock, so.. */
657 		old = find_inode_fast(sb, head, ino);
658 		if (!old) {
659 			inode->i_ino = ino;
660 			inodes_stat.nr_inodes++;
661 			list_add(&inode->i_list, &inode_in_use);
662 			list_add(&inode->i_sb_list, &sb->s_inodes);
663 			hlist_add_head(&inode->i_hash, head);
664 			inode->i_state = I_LOCK|I_NEW;
665 			spin_unlock(&inode_lock);
666 
667 			/* Return the locked inode with I_NEW set, the
668 			 * caller is responsible for filling in the contents
669 			 */
670 			return inode;
671 		}
672 
673 		/*
674 		 * Uhhuh, somebody else created the same inode under
675 		 * us. Use the old inode instead of the one we just
676 		 * allocated.
677 		 */
678 		__iget(old);
679 		spin_unlock(&inode_lock);
680 		destroy_inode(inode);
681 		inode = old;
682 		wait_on_inode(inode);
683 	}
684 	return inode;
685 }
686 
687 static inline unsigned long hash(struct super_block *sb, unsigned long hashval)
688 {
689 	unsigned long tmp;
690 
691 	tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
692 			L1_CACHE_BYTES;
693 	tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS);
694 	return tmp & I_HASHMASK;
695 }
696 
697 /**
698  *	iunique - get a unique inode number
699  *	@sb: superblock
700  *	@max_reserved: highest reserved inode number
701  *
702  *	Obtain an inode number that is unique on the system for a given
703  *	superblock. This is used by file systems that have no natural
704  *	permanent inode numbering system. An inode number is returned that
705  *	is higher than the reserved limit but unique.
706  *
707  *	BUGS:
708  *	With a large number of inodes live on the file system this function
709  *	currently becomes quite slow.
710  */
711 ino_t iunique(struct super_block *sb, ino_t max_reserved)
712 {
713 	static ino_t counter;
714 	struct inode *inode;
715 	struct hlist_head * head;
716 	ino_t res;
717 	spin_lock(&inode_lock);
718 retry:
719 	if (counter > max_reserved) {
720 		head = inode_hashtable + hash(sb,counter);
721 		res = counter++;
722 		inode = find_inode_fast(sb, head, res);
723 		if (!inode) {
724 			spin_unlock(&inode_lock);
725 			return res;
726 		}
727 	} else {
728 		counter = max_reserved + 1;
729 	}
730 	goto retry;
731 
732 }
733 
734 EXPORT_SYMBOL(iunique);
735 
736 struct inode *igrab(struct inode *inode)
737 {
738 	spin_lock(&inode_lock);
739 	if (!(inode->i_state & (I_FREEING|I_WILL_FREE)))
740 		__iget(inode);
741 	else
742 		/*
743 		 * Handle the case where s_op->clear_inode is not been
744 		 * called yet, and somebody is calling igrab
745 		 * while the inode is getting freed.
746 		 */
747 		inode = NULL;
748 	spin_unlock(&inode_lock);
749 	return inode;
750 }
751 
752 EXPORT_SYMBOL(igrab);
753 
754 /**
755  * ifind - internal function, you want ilookup5() or iget5().
756  * @sb:		super block of file system to search
757  * @head:       the head of the list to search
758  * @test:	callback used for comparisons between inodes
759  * @data:	opaque data pointer to pass to @test
760  * @wait:	if true wait for the inode to be unlocked, if false do not
761  *
762  * ifind() searches for the inode specified by @data in the inode
763  * cache. This is a generalized version of ifind_fast() for file systems where
764  * the inode number is not sufficient for unique identification of an inode.
765  *
766  * If the inode is in the cache, the inode is returned with an incremented
767  * reference count.
768  *
769  * Otherwise NULL is returned.
770  *
771  * Note, @test is called with the inode_lock held, so can't sleep.
772  */
773 static inline struct inode *ifind(struct super_block *sb,
774 		struct hlist_head *head, int (*test)(struct inode *, void *),
775 		void *data, const int wait)
776 {
777 	struct inode *inode;
778 
779 	spin_lock(&inode_lock);
780 	inode = find_inode(sb, head, test, data);
781 	if (inode) {
782 		__iget(inode);
783 		spin_unlock(&inode_lock);
784 		if (likely(wait))
785 			wait_on_inode(inode);
786 		return inode;
787 	}
788 	spin_unlock(&inode_lock);
789 	return NULL;
790 }
791 
792 /**
793  * ifind_fast - internal function, you want ilookup() or iget().
794  * @sb:		super block of file system to search
795  * @head:       head of the list to search
796  * @ino:	inode number to search for
797  *
798  * ifind_fast() searches for the inode @ino in the inode cache. This is for
799  * file systems where the inode number is sufficient for unique identification
800  * of an inode.
801  *
802  * If the inode is in the cache, the inode is returned with an incremented
803  * reference count.
804  *
805  * Otherwise NULL is returned.
806  */
807 static inline struct inode *ifind_fast(struct super_block *sb,
808 		struct hlist_head *head, unsigned long ino)
809 {
810 	struct inode *inode;
811 
812 	spin_lock(&inode_lock);
813 	inode = find_inode_fast(sb, head, ino);
814 	if (inode) {
815 		__iget(inode);
816 		spin_unlock(&inode_lock);
817 		wait_on_inode(inode);
818 		return inode;
819 	}
820 	spin_unlock(&inode_lock);
821 	return NULL;
822 }
823 
824 /**
825  * ilookup5_nowait - search for an inode in the inode cache
826  * @sb:		super block of file system to search
827  * @hashval:	hash value (usually inode number) to search for
828  * @test:	callback used for comparisons between inodes
829  * @data:	opaque data pointer to pass to @test
830  *
831  * ilookup5() uses ifind() to search for the inode specified by @hashval and
832  * @data in the inode cache. This is a generalized version of ilookup() for
833  * file systems where the inode number is not sufficient for unique
834  * identification of an inode.
835  *
836  * If the inode is in the cache, the inode is returned with an incremented
837  * reference count.  Note, the inode lock is not waited upon so you have to be
838  * very careful what you do with the returned inode.  You probably should be
839  * using ilookup5() instead.
840  *
841  * Otherwise NULL is returned.
842  *
843  * Note, @test is called with the inode_lock held, so can't sleep.
844  */
845 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
846 		int (*test)(struct inode *, void *), void *data)
847 {
848 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
849 
850 	return ifind(sb, head, test, data, 0);
851 }
852 
853 EXPORT_SYMBOL(ilookup5_nowait);
854 
855 /**
856  * ilookup5 - search for an inode in the inode cache
857  * @sb:		super block of file system to search
858  * @hashval:	hash value (usually inode number) to search for
859  * @test:	callback used for comparisons between inodes
860  * @data:	opaque data pointer to pass to @test
861  *
862  * ilookup5() uses ifind() to search for the inode specified by @hashval and
863  * @data in the inode cache. This is a generalized version of ilookup() for
864  * file systems where the inode number is not sufficient for unique
865  * identification of an inode.
866  *
867  * If the inode is in the cache, the inode lock is waited upon and the inode is
868  * returned with an incremented reference count.
869  *
870  * Otherwise NULL is returned.
871  *
872  * Note, @test is called with the inode_lock held, so can't sleep.
873  */
874 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
875 		int (*test)(struct inode *, void *), void *data)
876 {
877 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
878 
879 	return ifind(sb, head, test, data, 1);
880 }
881 
882 EXPORT_SYMBOL(ilookup5);
883 
884 /**
885  * ilookup - search for an inode in the inode cache
886  * @sb:		super block of file system to search
887  * @ino:	inode number to search for
888  *
889  * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache.
890  * This is for file systems where the inode number is sufficient for unique
891  * identification of an inode.
892  *
893  * If the inode is in the cache, the inode is returned with an incremented
894  * reference count.
895  *
896  * Otherwise NULL is returned.
897  */
898 struct inode *ilookup(struct super_block *sb, unsigned long ino)
899 {
900 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
901 
902 	return ifind_fast(sb, head, ino);
903 }
904 
905 EXPORT_SYMBOL(ilookup);
906 
907 /**
908  * iget5_locked - obtain an inode from a mounted file system
909  * @sb:		super block of file system
910  * @hashval:	hash value (usually inode number) to get
911  * @test:	callback used for comparisons between inodes
912  * @set:	callback used to initialize a new struct inode
913  * @data:	opaque data pointer to pass to @test and @set
914  *
915  * This is iget() without the read_inode() portion of get_new_inode().
916  *
917  * iget5_locked() uses ifind() to search for the inode specified by @hashval
918  * and @data in the inode cache and if present it is returned with an increased
919  * reference count. This is a generalized version of iget_locked() for file
920  * systems where the inode number is not sufficient for unique identification
921  * of an inode.
922  *
923  * If the inode is not in cache, get_new_inode() is called to allocate a new
924  * inode and this is returned locked, hashed, and with the I_NEW flag set. The
925  * file system gets to fill it in before unlocking it via unlock_new_inode().
926  *
927  * Note both @test and @set are called with the inode_lock held, so can't sleep.
928  */
929 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
930 		int (*test)(struct inode *, void *),
931 		int (*set)(struct inode *, void *), void *data)
932 {
933 	struct hlist_head *head = inode_hashtable + hash(sb, hashval);
934 	struct inode *inode;
935 
936 	inode = ifind(sb, head, test, data, 1);
937 	if (inode)
938 		return inode;
939 	/*
940 	 * get_new_inode() will do the right thing, re-trying the search
941 	 * in case it had to block at any point.
942 	 */
943 	return get_new_inode(sb, head, test, set, data);
944 }
945 
946 EXPORT_SYMBOL(iget5_locked);
947 
948 /**
949  * iget_locked - obtain an inode from a mounted file system
950  * @sb:		super block of file system
951  * @ino:	inode number to get
952  *
953  * This is iget() without the read_inode() portion of get_new_inode_fast().
954  *
955  * iget_locked() uses ifind_fast() to search for the inode specified by @ino in
956  * the inode cache and if present it is returned with an increased reference
957  * count. This is for file systems where the inode number is sufficient for
958  * unique identification of an inode.
959  *
960  * If the inode is not in cache, get_new_inode_fast() is called to allocate a
961  * new inode and this is returned locked, hashed, and with the I_NEW flag set.
962  * The file system gets to fill it in before unlocking it via
963  * unlock_new_inode().
964  */
965 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
966 {
967 	struct hlist_head *head = inode_hashtable + hash(sb, ino);
968 	struct inode *inode;
969 
970 	inode = ifind_fast(sb, head, ino);
971 	if (inode)
972 		return inode;
973 	/*
974 	 * get_new_inode_fast() will do the right thing, re-trying the search
975 	 * in case it had to block at any point.
976 	 */
977 	return get_new_inode_fast(sb, head, ino);
978 }
979 
980 EXPORT_SYMBOL(iget_locked);
981 
982 /**
983  *	__insert_inode_hash - hash an inode
984  *	@inode: unhashed inode
985  *	@hashval: unsigned long value used to locate this object in the
986  *		inode_hashtable.
987  *
988  *	Add an inode to the inode hash for this superblock.
989  */
990 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
991 {
992 	struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
993 	spin_lock(&inode_lock);
994 	hlist_add_head(&inode->i_hash, head);
995 	spin_unlock(&inode_lock);
996 }
997 
998 EXPORT_SYMBOL(__insert_inode_hash);
999 
1000 /**
1001  *	remove_inode_hash - remove an inode from the hash
1002  *	@inode: inode to unhash
1003  *
1004  *	Remove an inode from the superblock.
1005  */
1006 void remove_inode_hash(struct inode *inode)
1007 {
1008 	spin_lock(&inode_lock);
1009 	hlist_del_init(&inode->i_hash);
1010 	spin_unlock(&inode_lock);
1011 }
1012 
1013 EXPORT_SYMBOL(remove_inode_hash);
1014 
1015 /*
1016  * Tell the filesystem that this inode is no longer of any interest and should
1017  * be completely destroyed.
1018  *
1019  * We leave the inode in the inode hash table until *after* the filesystem's
1020  * ->delete_inode completes.  This ensures that an iget (such as nfsd might
1021  * instigate) will always find up-to-date information either in the hash or on
1022  * disk.
1023  *
1024  * I_FREEING is set so that no-one will take a new reference to the inode while
1025  * it is being deleted.
1026  */
1027 void generic_delete_inode(struct inode *inode)
1028 {
1029 	struct super_operations *op = inode->i_sb->s_op;
1030 
1031 	list_del_init(&inode->i_list);
1032 	list_del_init(&inode->i_sb_list);
1033 	inode->i_state|=I_FREEING;
1034 	inodes_stat.nr_inodes--;
1035 	spin_unlock(&inode_lock);
1036 
1037 	security_inode_delete(inode);
1038 
1039 	if (op->delete_inode) {
1040 		void (*delete)(struct inode *) = op->delete_inode;
1041 		if (!is_bad_inode(inode))
1042 			DQUOT_INIT(inode);
1043 		/* Filesystems implementing their own
1044 		 * s_op->delete_inode are required to call
1045 		 * truncate_inode_pages and clear_inode()
1046 		 * internally */
1047 		delete(inode);
1048 	} else {
1049 		truncate_inode_pages(&inode->i_data, 0);
1050 		clear_inode(inode);
1051 	}
1052 	spin_lock(&inode_lock);
1053 	hlist_del_init(&inode->i_hash);
1054 	spin_unlock(&inode_lock);
1055 	wake_up_inode(inode);
1056 	if (inode->i_state != I_CLEAR)
1057 		BUG();
1058 	destroy_inode(inode);
1059 }
1060 
1061 EXPORT_SYMBOL(generic_delete_inode);
1062 
1063 static void generic_forget_inode(struct inode *inode)
1064 {
1065 	struct super_block *sb = inode->i_sb;
1066 
1067 	if (!hlist_unhashed(&inode->i_hash)) {
1068 		if (!(inode->i_state & (I_DIRTY|I_LOCK)))
1069 			list_move(&inode->i_list, &inode_unused);
1070 		inodes_stat.nr_unused++;
1071 		if (!sb || (sb->s_flags & MS_ACTIVE)) {
1072 			spin_unlock(&inode_lock);
1073 			return;
1074 		}
1075 		inode->i_state |= I_WILL_FREE;
1076 		spin_unlock(&inode_lock);
1077 		write_inode_now(inode, 1);
1078 		spin_lock(&inode_lock);
1079 		inode->i_state &= ~I_WILL_FREE;
1080 		inodes_stat.nr_unused--;
1081 		hlist_del_init(&inode->i_hash);
1082 	}
1083 	list_del_init(&inode->i_list);
1084 	list_del_init(&inode->i_sb_list);
1085 	inode->i_state |= I_FREEING;
1086 	inodes_stat.nr_inodes--;
1087 	spin_unlock(&inode_lock);
1088 	if (inode->i_data.nrpages)
1089 		truncate_inode_pages(&inode->i_data, 0);
1090 	clear_inode(inode);
1091 	wake_up_inode(inode);
1092 	destroy_inode(inode);
1093 }
1094 
1095 /*
1096  * Normal UNIX filesystem behaviour: delete the
1097  * inode when the usage count drops to zero, and
1098  * i_nlink is zero.
1099  */
1100 void generic_drop_inode(struct inode *inode)
1101 {
1102 	if (!inode->i_nlink)
1103 		generic_delete_inode(inode);
1104 	else
1105 		generic_forget_inode(inode);
1106 }
1107 
1108 EXPORT_SYMBOL_GPL(generic_drop_inode);
1109 
1110 /*
1111  * Called when we're dropping the last reference
1112  * to an inode.
1113  *
1114  * Call the FS "drop()" function, defaulting to
1115  * the legacy UNIX filesystem behaviour..
1116  *
1117  * NOTE! NOTE! NOTE! We're called with the inode lock
1118  * held, and the drop function is supposed to release
1119  * the lock!
1120  */
1121 static inline void iput_final(struct inode *inode)
1122 {
1123 	struct super_operations *op = inode->i_sb->s_op;
1124 	void (*drop)(struct inode *) = generic_drop_inode;
1125 
1126 	if (op && op->drop_inode)
1127 		drop = op->drop_inode;
1128 	drop(inode);
1129 }
1130 
1131 /**
1132  *	iput	- put an inode
1133  *	@inode: inode to put
1134  *
1135  *	Puts an inode, dropping its usage count. If the inode use count hits
1136  *	zero, the inode is then freed and may also be destroyed.
1137  *
1138  *	Consequently, iput() can sleep.
1139  */
1140 void iput(struct inode *inode)
1141 {
1142 	if (inode) {
1143 		struct super_operations *op = inode->i_sb->s_op;
1144 
1145 		BUG_ON(inode->i_state == I_CLEAR);
1146 
1147 		if (op && op->put_inode)
1148 			op->put_inode(inode);
1149 
1150 		if (atomic_dec_and_lock(&inode->i_count, &inode_lock))
1151 			iput_final(inode);
1152 	}
1153 }
1154 
1155 EXPORT_SYMBOL(iput);
1156 
1157 /**
1158  *	bmap	- find a block number in a file
1159  *	@inode: inode of file
1160  *	@block: block to find
1161  *
1162  *	Returns the block number on the device holding the inode that
1163  *	is the disk block number for the block of the file requested.
1164  *	That is, asked for block 4 of inode 1 the function will return the
1165  *	disk block relative to the disk start that holds that block of the
1166  *	file.
1167  */
1168 sector_t bmap(struct inode * inode, sector_t block)
1169 {
1170 	sector_t res = 0;
1171 	if (inode->i_mapping->a_ops->bmap)
1172 		res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1173 	return res;
1174 }
1175 
1176 EXPORT_SYMBOL(bmap);
1177 
1178 /**
1179  *	update_atime	-	update the access time
1180  *	@inode: inode accessed
1181  *
1182  *	Update the accessed time on an inode and mark it for writeback.
1183  *	This function automatically handles read only file systems and media,
1184  *	as well as the "noatime" flag and inode specific "noatime" markers.
1185  */
1186 void update_atime(struct inode *inode)
1187 {
1188 	struct timespec now;
1189 
1190 	if (IS_NOATIME(inode))
1191 		return;
1192 	if (IS_NODIRATIME(inode) && S_ISDIR(inode->i_mode))
1193 		return;
1194 	if (IS_RDONLY(inode))
1195 		return;
1196 
1197 	now = current_fs_time(inode->i_sb);
1198 	if (!timespec_equal(&inode->i_atime, &now)) {
1199 		inode->i_atime = now;
1200 		mark_inode_dirty_sync(inode);
1201 	}
1202 }
1203 
1204 EXPORT_SYMBOL(update_atime);
1205 
1206 /**
1207  *	inode_update_time	-	update mtime and ctime time
1208  *	@inode: inode accessed
1209  *	@ctime_too: update ctime too
1210  *
1211  *	Update the mtime time on an inode and mark it for writeback.
1212  *	When ctime_too is specified update the ctime too.
1213  */
1214 
1215 void inode_update_time(struct inode *inode, int ctime_too)
1216 {
1217 	struct timespec now;
1218 	int sync_it = 0;
1219 
1220 	if (IS_NOCMTIME(inode))
1221 		return;
1222 	if (IS_RDONLY(inode))
1223 		return;
1224 
1225 	now = current_fs_time(inode->i_sb);
1226 	if (!timespec_equal(&inode->i_mtime, &now))
1227 		sync_it = 1;
1228 	inode->i_mtime = now;
1229 
1230 	if (ctime_too) {
1231 		if (!timespec_equal(&inode->i_ctime, &now))
1232 			sync_it = 1;
1233 		inode->i_ctime = now;
1234 	}
1235 	if (sync_it)
1236 		mark_inode_dirty_sync(inode);
1237 }
1238 
1239 EXPORT_SYMBOL(inode_update_time);
1240 
1241 int inode_needs_sync(struct inode *inode)
1242 {
1243 	if (IS_SYNC(inode))
1244 		return 1;
1245 	if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1246 		return 1;
1247 	return 0;
1248 }
1249 
1250 EXPORT_SYMBOL(inode_needs_sync);
1251 
1252 /*
1253  *	Quota functions that want to walk the inode lists..
1254  */
1255 #ifdef CONFIG_QUOTA
1256 
1257 /* Function back in dquot.c */
1258 int remove_inode_dquot_ref(struct inode *, int, struct list_head *);
1259 
1260 void remove_dquot_ref(struct super_block *sb, int type,
1261 			struct list_head *tofree_head)
1262 {
1263 	struct inode *inode;
1264 
1265 	if (!sb->dq_op)
1266 		return;	/* nothing to do */
1267 	spin_lock(&inode_lock);	/* This lock is for inodes code */
1268 
1269 	/*
1270 	 * We don't have to lock against quota code - test IS_QUOTAINIT is
1271 	 * just for speedup...
1272 	 */
1273 	list_for_each_entry(inode, &sb->s_inodes, i_sb_list)
1274 		if (!IS_NOQUOTA(inode))
1275 			remove_inode_dquot_ref(inode, type, tofree_head);
1276 
1277 	spin_unlock(&inode_lock);
1278 }
1279 
1280 #endif
1281 
1282 int inode_wait(void *word)
1283 {
1284 	schedule();
1285 	return 0;
1286 }
1287 
1288 /*
1289  * If we try to find an inode in the inode hash while it is being
1290  * deleted, we have to wait until the filesystem completes its
1291  * deletion before reporting that it isn't found.  This function waits
1292  * until the deletion _might_ have completed.  Callers are responsible
1293  * to recheck inode state.
1294  *
1295  * It doesn't matter if I_LOCK is not set initially, a call to
1296  * wake_up_inode() after removing from the hash list will DTRT.
1297  *
1298  * This is called with inode_lock held.
1299  */
1300 static void __wait_on_freeing_inode(struct inode *inode)
1301 {
1302 	wait_queue_head_t *wq;
1303 	DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK);
1304 	wq = bit_waitqueue(&inode->i_state, __I_LOCK);
1305 	prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1306 	spin_unlock(&inode_lock);
1307 	schedule();
1308 	finish_wait(wq, &wait.wait);
1309 	spin_lock(&inode_lock);
1310 }
1311 
1312 void wake_up_inode(struct inode *inode)
1313 {
1314 	/*
1315 	 * Prevent speculative execution through spin_unlock(&inode_lock);
1316 	 */
1317 	smp_mb();
1318 	wake_up_bit(&inode->i_state, __I_LOCK);
1319 }
1320 
1321 static __initdata unsigned long ihash_entries;
1322 static int __init set_ihash_entries(char *str)
1323 {
1324 	if (!str)
1325 		return 0;
1326 	ihash_entries = simple_strtoul(str, &str, 0);
1327 	return 1;
1328 }
1329 __setup("ihash_entries=", set_ihash_entries);
1330 
1331 /*
1332  * Initialize the waitqueues and inode hash table.
1333  */
1334 void __init inode_init_early(void)
1335 {
1336 	int loop;
1337 
1338 	/* If hashes are distributed across NUMA nodes, defer
1339 	 * hash allocation until vmalloc space is available.
1340 	 */
1341 	if (hashdist)
1342 		return;
1343 
1344 	inode_hashtable =
1345 		alloc_large_system_hash("Inode-cache",
1346 					sizeof(struct hlist_head),
1347 					ihash_entries,
1348 					14,
1349 					HASH_EARLY,
1350 					&i_hash_shift,
1351 					&i_hash_mask,
1352 					0);
1353 
1354 	for (loop = 0; loop < (1 << i_hash_shift); loop++)
1355 		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1356 }
1357 
1358 void __init inode_init(unsigned long mempages)
1359 {
1360 	int loop;
1361 
1362 	/* inode slab cache */
1363 	inode_cachep = kmem_cache_create("inode_cache", sizeof(struct inode),
1364 				0, SLAB_RECLAIM_ACCOUNT|SLAB_PANIC, init_once, NULL);
1365 	set_shrinker(DEFAULT_SEEKS, shrink_icache_memory);
1366 
1367 	/* Hash may have been set up in inode_init_early */
1368 	if (!hashdist)
1369 		return;
1370 
1371 	inode_hashtable =
1372 		alloc_large_system_hash("Inode-cache",
1373 					sizeof(struct hlist_head),
1374 					ihash_entries,
1375 					14,
1376 					0,
1377 					&i_hash_shift,
1378 					&i_hash_mask,
1379 					0);
1380 
1381 	for (loop = 0; loop < (1 << i_hash_shift); loop++)
1382 		INIT_HLIST_HEAD(&inode_hashtable[loop]);
1383 }
1384 
1385 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1386 {
1387 	inode->i_mode = mode;
1388 	if (S_ISCHR(mode)) {
1389 		inode->i_fop = &def_chr_fops;
1390 		inode->i_rdev = rdev;
1391 	} else if (S_ISBLK(mode)) {
1392 		inode->i_fop = &def_blk_fops;
1393 		inode->i_rdev = rdev;
1394 	} else if (S_ISFIFO(mode))
1395 		inode->i_fop = &def_fifo_fops;
1396 	else if (S_ISSOCK(mode))
1397 		inode->i_fop = &bad_sock_fops;
1398 	else
1399 		printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n",
1400 		       mode);
1401 }
1402 EXPORT_SYMBOL(init_special_inode);
1403